Pinned
Gradium is out of stealth to solve voice. We raised $70M and after only 3 months we’re releasing our transcription and synthesis products to power the next generation of voice AI.
00:00
Gradium
139 posts
Gradium
@GradiumAI
The voice layer for modern apps and agents. Real-time, scalable voice APIs: TTS, STT, turn-taking & voice cloning. Devs: build → gradium.ai/#models
Joined September 2025
- AIEWF next week. We'll be at booth U-G8 with @pipecat_ai, and our CEO @neilzegh is giving 3 talks: → Your voice agent is just a walkie-talkie → Voice is the universal interface, w/ @kwindla → Everyone gets a digital clone
00:00 - Today we launch stt-translate and s2s-translate: real-time speech-to-text and speech-to-speech translation. They compete with gemini-3.5-live-translate and gpt-realtime-translate on latency and quality, while allowing you to speak in any voice from our catalog or one you clone.
- Gradium STT and TTS now powers @adoptbuddy , the emotional companion robot from @BlueFrogRobotic, deployed in schools, hospitals, and senior care.
- Our on-device TTS model Phonon out there crushing it on TTFA and WER.Long flights always give me more ideas to think about what's missing around us. Few prompts later, here's Scribble Story. On-device fully local pipeline to convert scribblings into a short story you can listen to. Using @GradiumAI Phonon and @Alibaba_Qwen
00:00 - 60+ new voices live in the Gradium catalogue. English, Spanish, French, German, and Portuguese, with eight regional accents across them. gradium.ai
GIF - ¡Hola Barcelona☀️Next speaker are @ConstanceGriso and Timothé from @GradiumAI talking about Phonon, their on device model Really cool demo
- We upgraded Gradium TTS for the cases voice agents can't get wrong: phone numbers, codes, email addresses read back right the first time. Couple of examples: English: 97% on emails, top of the field. French: leads every competitor we benchmarked. Samples + methodology →
00:00 - In this joint work with @kyutai_labs, we design a reward model for conversational dynamics to teach full-duplex models how a human behaves in conversation, using cues to know when to interrupt, backchannel or stay silent.New paper: Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models We use RL to post-train speech models (Moshi and PersonaPlex) to talk more like a human: to know when to respond, when to wait, and when to nod along with “yeah”s and “okay”s when listening.
00:00 - We'll be at @VivaTech next week showcasing our models. Come find us at Booth 7.2 | 2F13 with @awscloud all week, and on the @LaFrenchTech booth on Wednesday. @neilzegh is giving two talks: Wed 17th, 5:20pm, @nvidia Stage 1 and on Fri, 10am, Théâtre AWS
- Learn how to build an audiobook voice agent using Gradium and @pipecat_ai Gradium's TTS handles the narration and Pipecat's built-in WebRTC transport delivers the audio to the browser.
00:00 - Reasoning LLMs typically take 2-3 seconds to start emitting tokens. In a voice agent, that's 2-3 seconds of silence after the user finishes speaking. The @MiniMax_AI team just shipped a community contribution to Gradbot with two models running in parallel. MiniMax-M2-her
- A full house at the @joinhexa office in Paris yesterday. Our CTO @olivierteboul joined the discussion by sharing why low latency matters for voice agents and how Gradium models support enterprise use cases for voice AI.
